Article

Low Dimensionality in Gene Expression Data

Enables the Accurate Extraction of Transcriptional
Programs from Shallow Sequencing
Graphical Abstract Authors
Graham Heimberg, Rajat Bhatnagar,
Hana El-Samad, Matt Thomson

Correspondence
hana.el-samad@ucsf.edu (H.E.-S.),
matthew.thomson@ucsf.edu (M.T.)

In Brief
We develop a mathematical framework
that delineates how parameters such as
read depth and sample number influence
the error in transcriptional program
extraction from mRNA-sequencing data.
Our analyses reveal that gene expression
modularity facilitates low error at
surprisingly low read depths, arguing that
increased multiplexing of shallow
sequencing experiments is a viable
approach for applications such as single-
cell profiling of entire tumors.

Highlights
d Mathematical model reveals impact of mRNA-seq read depth
on gene expression analysis

d Modularity in gene expression facilitates robust

transcriptional program extraction

d Model suggests dramatic increases in sample multiplexing

for many applications

d Read depth calculator determines parameters for optimal

experimental design

Heimberg et al., 2016, Cell Systems 2, 239250

April 27, 2016 2016 The Authors
http://dx.doi.org/10.1016/j.cels.2016.04.001
Cell Systems

Article

Low Dimensionality in Gene Expression Data

Enables the Accurate Extraction of Transcriptional
Programs from Shallow Sequencing
Graham Heimberg,1,2,3,4,5 Rajat Bhatnagar,1,3,5 Hana El-Samad,1,3,* and Matt Thomson3,4,*
1Department of Biochemistry and Biophysics, California Institute for Quantitative Biosciences, University of California, San Francisco, San

Francisco, CA 94158, USA

2Integrative Program in Quantitative Biology, University of California, San Francisco, San Francisco, CA 94158, USA
3Center for Systems and Synthetic Biology, University of California, San Francisco, San Francisco, CA 94158, USA
4Department of Cellular and Molecular Pharmacology, University of California, San Francisco, San Francisco, CA 94158, USA
5Co-first author

*Correspondence: hana.el-samad@ucsf.edu (H.E.-S.), matthew.thomson@ucsf.edu (M.T.)

http://dx.doi.org/10.1016/j.cels.2016.04.001

SUMMARY Not all biological questions require such extreme technical

A tradeoff between precision and throughput con-
strains all biological measurements, including seq-
uencing-based technologies. Here, we develop a
mathematical framework that defines this tradeoff
between mRNA-sequencing depth and error in the
extraction of biological information. We find that
transcriptional programs can be reproducibly identi-
fied at 1% of conventional read depths. We demon-
strate that this resilience to noise of shallow
sequencing derives from a natural property, low
dimensionality, which is a fundamental feature of
gene expression data. Accordingly, our conclusions
hold for !350 single-cell and bulk gene expression
datasets across yeast, mouse, and human. In total,
our approach provides quantitative guidelines for
the choice of sequencing depth necessary to achieve
a desired level of analytical resolution. We codify
these guidelines in an open-source read depth calcu-
lator. This work demonstrates that the structure
inherent in biological networks can be productively
exploited to increase measurement throughput, an
idea that is now common in many branches of sci-
ence, such as image processing.
INTRODUCTION
All measurements, including biological measurements, contain a
tradeoff between precision and throughput. In sequencing-based
measurements like mRNA-sequencing (mRNA-seq), precision is
determined largely by the sequencing depth applied to individual
samples. At high sequencing depth, mRNA-seq can detect subtle
changes in gene expression including the expression of rare
splice variants or quantitative modulations in transcript abun-
dance. However, such precision comes at a cost, and sequencing
transcripts from 10,000 single cells at deep sequencing coverage
(106 reads per cell) currently requires 2 weeks of sequencing on an
Illumina HiSeq 4000.
sensitivity. For example, a catalog of human cell types and
the transcriptional programs that define them can potentially
be generated by querying the general transcriptional state of
single cells (Trapnell, 2015). In principle, theoretical and
computational methods could elucidate the tradeoff between
sequencing depth and granularity of the information that can
be accurately extracted from samples. Accordingly, opti-
mizing this tradeoff based on the granularity required by the
biological question at hand would yield significant increases
in the scale at which mRNA-seq can be applied, facilitating
applications such as drug screening and whole-organ or tu-
mor profiling.
The modern engineering discipline of signal processing has
demonstrated that structural properties of natural signals can
often be exploited to enable new classes of low cost measure-
ments. The central insight is that many natural signals are
effectively low dimensional. Geometrically, this means
that these signals lie on a noisy, low-dimensional manifold
embedded in the observed, high-dimensional measurement
space. Equivalently, this property indicates that there is a
basis representation in which these signals can be accurately
captured by a small number of basis vectors relative to the orig-
inal measurement dimension (Donoho, 2006; Candes et al.,
2006; Hinton and Salakhutdinov, 2006). Modern algorithms
exploit the fact that the number of measurements required to
reconstruct a low-dimensional signal can be far fewer than the
apparent number of degrees of freedom. For example, in images
of natural scenes, correlations between neighboring pixels
induce an effective low dimensionality that allows high-accuracy
image reconstruction even in the presence of considerable mea-
surement noise such as point defects in many camera pixels
(Duarte et al., 2008).
Like natural images, it has long been appreciated that biolog-
ical systems contain structural features that can lead to an
effective low dimensionality in data. Most notably, genes are
commonly co-regulated within transcriptional modules; this pro-
duces covariation in the expression of many genes (Eisen et al.,
1998; Segal et al., 2003; Bergmann et al., 2003). The widespread
presence of such modules indicates that the natural dimension-
ality of gene expression is determined not by the number of genes
in the genome but by the number of regulatory modules. By

Cell Systems 2, 239250, April 27, 2016 2016 The Authors 239
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Figure 1. A Mathematical Model Reveals Factors Determining the Performance of Shallow mRNA-Seq
(A) mRNA-seq throughput as a function of sequencing depth per sample for a fixed sequencing capacity.
(B) Unsupervised learning techniques are used to identify transcriptional programs. We ask when and why shallow mRNA-seq can accurately identify tran-
scriptional programs.
(C) Decreasing sequencing depth adds measurement noise to the transcriptional programs identified by unsupervised learning. Our approach reveals that
dominant programs, defined as those that explain relatively large variances in the data, are tolerant to measurement noise.

analogy to signal processing, this natural structure suggests that
the lower effective dimensionality present in gene expression
data can be exploited to make accurate, inexpensive measure-
ments that are not degraded by noise. But when, and at what
error tradeoff, can low dimensionality be leveraged to enable
low-cost, high-information-content biological measurements?
Here, inspired by these developments in signal processing,
we establish a mathematical framework that addresses the
impact of reducing coverage depth, and hence increasing mea-
surement noise, on the reconstruction of transcriptional regula-
tory programs from mRNA-seq data. Our framework reveals
that shallow mRNA-seq, which has been proposed to in-
crease mRNA-seq throughput by reducing sequencing depth
in individual samples (Jaitin et al., 2014; Pollen et al., 2014; Klie-
benstein, 2012) (Figure 1A), can be applied generally to many
bulk and single-cell mRNA-seq experiments. By investigating
the fundamental limits of shallow mRNA-seq, we define the
conditions under which it has utility and complements deep
sequencing.
Our analysis reveals that the dominance of a transcriptional
program, quantified by the fraction of the variance it explains
in the dataset, determines the read depth required to accu-
rately extract it. We demonstrate that common bioinformatic
analyses can be performed at 1% of traditional sequencing
depths with little loss in inferred biological information at the
level of transcriptional programs. We also introduce a simple
read depth calculator that determines optimal experimental
parameters to achieve a desired analytical accuracy. Our
framework and computational results highlight the effective
low dimensionality of gene expression, commonly caused by
co-regulation of genes, as both a fundamental feature of
biological data and a major underpinning of biological sig-
nals tolerance to measurement noise (Figures 1B and 1C). Un-
derstanding the fundamental limits and tradeoffs involved in
extracting information from mRNA-seq data will guide re-
searchers in designing large-scale bulk mRNA-seq experi-
ments and analyzing single-cell data where transcript coverage
is inherently low.

240 Cell Systems 2, 239250, April 27, 2016

RESULTS
Statistical Properties of Gene Expression Data
Determine the Accuracy of Principal Component
Analysis at Low Read Depth
To delineate the impact of sequencing depth on the analysis of
mRNA-seq data, we developed a mathematical framework that
models the performance of a common bioinformatics tech-
nique, transcriptional program identification, at low sequencing
depth. We focus on transcriptional program identification as it
is central in many analyses including gene set analysis, network
reconstruction (Holter et al., 2001; Bonneau, 2008), and cancer
classification (Alon et al., 1999; Shai et al., 2003; Patel et al.,
2014), as well as the analysis of single-cell mRNA-seq data.
Our model defines exactly how reductions in read depth
corrupt the extracted transcriptional programs and determines
the precise depth required to recover them with a desired
accuracy.
Our analysis focuses on the identification of transcriptional
programs from mRNA-seq data through principal component
analysis (PCA), because of its prevalence in gene expression
analysis (Alter et al., 2000; Ringner, 2008) and its fundamental
similarities to other commonly used methods. A recent review
called PCA the most widely used method for unsupervised clus-
tering and noted that it has already been successfully applied in
many single-cell genomics contexts (Trapnell, 2015). Addition-
ally, research in the computer science community over the
last decade has shown that many other unsupervised learning
methods, including k-means, spectral clustering, and Locally
Linear Embedding, are naturally related to PCA or its generaliza-
tion, Kernel PCA (Ding and He, 2004; Ng et al., 2001; Ham et al.,
2004; Bengio et al., 2004). Because of the deep connection
between PCA and other unsupervised learning techniques,
we expect that our conclusions in this section will extend to other
methods of analysis (and we provide such parallel analysis in the
Supplemental Information). Here, we focus on PCA because the
well-defined theory behind it provides a unique opportunity to
understand, analytically, the factors that determine the robust-
ness of program identification to low-coverage sequencing
noise.
PCA identifies transcriptional programs by extracting groups
of genes that covary across a set of samples. Covarying genes
are grouped into a gene expression vector known as a principal
component. Principal components are weighted by their relative
importance in capturing the gene expression variation that oc-
curs in the underlying data. Decreasing sequencing depth intro-
duces measurement noise into the gene expression data and
corrupts the extracted principal components.
If the transcriptional programs obtained from shallow mRNA-
seq data and deep mRNA-seq data are similar, then we can
accurately perform many gene expression analyses at low depth
while collecting data in much higher throughput (Figure 1). We
therefore developed a mathematical model that quantifies how
the principal components computed at low and high sequencing
depths differ. The model reveals that performance of transcrip-
tional program extraction at low read depth is specific to the da-
taset and even the program itself. It is the dominant transcrip-
tional programs, which capture most variance, that are the
most stable.
Formally, the principal components are defined as the eigen-
vectors of the gene expression covariance matrix, and the prin-
cipal values li are the associated eigenvalues that equal the
variance of the data projected onto the component (Alter
et al., 2000; Holter et al., 2001). We use perturbation theory
to model how the eigenvectors of the gene expression covari-
ance matrix change when measurement noise is added (Stew-
art and Sun, 1990; Shankar, 2012). We perform our analysis
in units of normalized read counts for conceptual clarity (or
normalized transcript counts where appropriate), but an iden-
tical analysis and error equation can be derived in FPKM units
through a simple rescaling. The principal component error is
defined as the deviation between the deep (pci) and shallow
dpci principal components,
v
u 0 " # 12
uX pcT C b $ C pcj
u i
kpci $ pci k z
d t @ A (Equation 1)
jsi
li $ lj
where C and C ^ are the covariance matrices obtained from deep
and shallow mRNA-seq data, respectively. Equation 1 can be
used to model the impact of shallow sequencing on any given
mRNA-seq dataset. Moreover, qualitative analysis of the equa-
tion reveals the key factors that determine whether low depth
profiling will accurately identify transcriptional programs. As
expected, this equation indicates that the principal component
error depends on generic features including read depth and
sample number, as these affect the difference between the
shallow and deep covariance matrices in the numerator of Equa-
tion 1 (see the Supplemental Information, section 2.1). However,
Equation 1 also reveals that the principal component error de-
pends on a system-specific property: the relative magnitude of
the principal values (captured by li $ lj). Since the principal
values correspond to the variance in the data along a principal
component, this term quantifies whether the information in the
gene expression data is concentrated among a few transcrip-
tional programs. When genes covary along a small number of
principal axes, the dataset has an effective low dimensionality,
i.e., the data are concentrated on a low-dimensional sub-space,
and transcriptional programs can be extracted even in the pres-
ence of sequencing noise.
Mouse Tissues Can Be Distinguished at Low Depth in
Bulk mRNA-Seq Samples
To understand the implications of this result in the context of
an established mRNA-seq dataset, we applied Equation 1 to
a subset of the mouse ENCODE data that uses deep mRNA-
seq (>107 reads per sample) to profile gene expression of
19 different mouse tissues with a biological replicate (Shen
et al., 2012) (see the Experimental Procedures). The analysis
revealed that the leading, dominant transcriptional programs
could be extracted with <1% of the studies original read
depth. Specifically, the first three principal components could
be recovered with >80% accuracy (i.e., an error of 1 $ 0.8 =
20%) with just 55,000 reads per experiment (Figures 2A and
S1A). To reach 80% accuracy for all of the first nine principal
components, only 145,000 reads were needed (Figure S1B).
Increasing read depth further had diminishing returns for prin-
cipal component accuracy. To increase the accuracy of the
Cell Systems 2, 239250, April 27, 2016 241
Figure 2. Transcriptional States of Mouse Tissues Are Distinguishable at Low Read Coverage
(A) Principal component error as a function of read depth for selected principal components for the Shen et al. (2012) data. For first three principal components,
1% of the traditional read depth is sufficient for achieving >80% accuracy. Improvements in error exhibit diminishing returns as read depth is increased. Less
dominant transcription programs (principal components 8 and 15 shown) are more sensitive to sequencing noise.
(B) Variance explained by transcriptional program (blue) and differences between principal values (green) of the Shen et al. (2012) data. The leading, dominant
transcriptional programs have principal values that are well separated from later principal values, suggesting that these should be more robust to measurement
noise.
(C) GSEA significance for the top ten terms of principal component two (top) and three (bottom) as a function of read depth. 32,000 reads are sufficient to recover
all top ten terms in the first three principal components. (Analysis for first principal component shown in Figure S1C.)
(D) Projection of a subset of the Shen et al. (2012) tissue data onto principal components two and three. The ellipses represent uncertainty at specific read depths.
Similar tissues lie close together. Transcriptional program two separates neural tissues from non-neural tissues while transcriptional program three distinguishes
tissues involved in hematopoiesis from other tissues. This is consistent with the GSEA of these transcriptional programs in (C).

first three principal components an additional 5% (from 80%
to 85%), 55% more reads were required. We confirmed these
analytical results by simulating shallow mRNA-seq through
direct sub-sampling of reads from the raw dataset (see the
Experimental Procedures).
Further, as predicted by Equation 1, the dominant principal
components were more robust to shallow sequencing noise
than the trailing, minor principal components. This is a direct
consequence of the fact that the leading principal values are
well separated from other principal values, while the trailing
values are spaced closely together. For instance, l1 is separated
from other principal values by at least l1 $ l2 = 5 3 10$6, more
than two orders of magnitude greater than the minimum separa-
tion of l25 from other principal values (1.5 3 10$8) (Figure 2B).
Therefore, the 25th principal component requires almost four
million reads, 140 times more than the first principal component,
to be recovered with the same 80% accuracy.
To explore whether the shallow principal components also
retained the same biological information as the programs
computed from deep mRNA-seq data, we compared results
from Gene Set Enrichment Analysis applied to shallow and
deep mRNA-seq data. At a read depth of 107 reads per sample,
the first three principal components have many significant func-
tional enrichments with the second and third principal compo-
nents enriched for neural and hematopoietic processes, respec-
tively (Figure 2C; see Figure S1C for first principal component).
These functional enrichments corroborate the separation seen
when the gene expression profiles from each tissue are pro-
jected onto the second and third principal components (see
the Experimental Procedures). Neural tissues (cerebellum,

242 Cell Systems 2, 239250, April 27, 2016

cortex, olfactory, and embryonic day 14.5 [E14.5] brain) project
along the second principal component while the hematopoietic
tissues (spleen, liver, thymus, bone marrow, and E14.5 liver)
project along the third principal component (Figure 2D).
The statistically significant enrichments of the first three prin-
cipal components persisted at low sequencing depths. At
<32,000 reads per sample, only 0.37% of the total reads, all
ten of the top gene sets for these principal components passed
our significance threshold of p < 10$4 (negative predictive value
and positive predictive value in Figures S1D and S1E). To put this
result in perspective, using only 32,000 reads per sample (corre-
sponding to PCA accuracies of 81%, 79%, and 75% for the first
three principal components, respectively) would allow a faithful
recapitulation of functional enrichments while still multiplexing
thousands of samples, rather than dozens, in a single Illumina
HiSeq sequencing lane. Additionally, this low number of reads
was still sufficient to separate the different cell types (Figure 2D).
We obtained similar results when working in FPKM units, sug-
gesting that the broad conclusions of our analysis are insensitive
to gene expression units (Figures S1F, S1G, and S1H).
Transcriptional States in Single Cells Are
Distinguishable with Less Than 1,000 Transcripts
per Cell
We wanted to explore whether shallow mRNA-seq could also
capture gene expression differences between individual single
cells within a heterogeneous tissue, arguably a more challenging
problem than distinguishing different bulk tissue samples. In
addition to the biological importance of quantifying variability
at the single-cell level, single-cell mRNA-seq data provide the
necessary context for analyzing the performance of shallow
sequencing for two reasons. First, single-cell mRNA-seq exper-
iments are inherently low-depth measurements as current ated with the first principal component could still be detected
methods can capture only a small fraction (!20%) (Shalek
et al., 2014) of the !300,000 transcripts (Velculescu et al.,
1999) typically contained in individual cells. Second, since ad-
vances in microfluidics (Macosko et al., 2015) now facilitate the
automated preparation of tens of thousands of individual cells
for single-cell mRNA-seq, sequencing requirements impose a
key bottleneck on the further scaling of single-cell throughput.
To probe the impact of sequencing depth reductions on sin-
gle-cell mRNA-seq data, we analyzed a dataset characterizing
3,005 single cells from the mouse cerebral cortex and hippocam-
pus (Zeisel et al., 2015) that were classified bioinformatically at
full sequencing depth (average of !15,000 unique transcripts
per cell) into nine different neural and non-neural cell types. In
addition to providing a rich biological context for analysis, this
dataset allows for a quantitative analysis of low-depth transcrip-
tional profiling as it incorporates molecular barcodes known as
unique molecular identifiers (UMIs) that enable the precise
counting of transcripts from each single cell. The Zeisel et al.
(2015) data therefore allowed us to analyze the impact of
sequencing depth reductions quantitatively in units of transcript
counts rather than in the less precise unit of raw sequencing
reads.
Similarly to the bulk tissue data, we found that leading prin-
cipal components in single cells could be reconstructed with a
small fraction of the total transcripts collected in the raw dataset.
We focused our analysis on three classes of cell typestwo
classes of pyramidal neurons with similar gene expression pro-
files and oligodendrocytesthat are transcriptionally distinct.
As the first three principal values were well separated from the
others (Figure S2A), Equation 1 estimated that the first three prin-
cipal components could be reconstructed with 11%, 22%, and
38% error, respectively, with just 1,000 transcripts per cell
(Figure 3A).
We confirmed this result computationally. With just 100 unique
transcripts, we were able to separate oligodendrocytes from
the two classes of pyramidal neurons with >90% accuracy.
With 1,000 unique transcripts per cell, we were able to distin-
guish pyramidal neurons of the hippocampus from those of cor-
tex with the same >90% accuracy (Figure 3B). The different
depths required to distinguish these subclasses of neural and
non-neural cell-types reflect the differing robustness of the cor-
responding principal components. The first principal component
captures a broad distinction between oligodendrocytes and py-
ramidal cell types (Figure 3C, left) and is the most robust to low
read depths. The third principal component captures a more
fine-grained distinction between pyramidal neurons but is less
robust than the first principal component at low read depth
and hence requires more coverage. This is consistent with bio-
logical intuition: more depth is required to distinguish between
pyramidal neural subtypes than between oligodendrocytes and
pyramidal neurons.
We next asked how contributions of individual genes to a prin-
cipal component change as a function of read depth. For every
principal component, we derived a null model consisting of the
distribution of the individual gene weightings, called loadings,
from a shuffled version of the data (see the Experimental Proce-
dures). Comparing the data to the null model, we found that at a
depth of !340 transcripts, >80% of genes significantly associ-
(Figures 3C and 3D; Experimental Procedures). At just 100 tran-
scripts per cell, we were still able to identify oligodendroycte
markers, such as myelin-associated oligodendrocyte basic pro-
tein (Mobp) and myelin-associated glycoprotein (Mag), as well as
neural markers, such as Neuronal differentiation 6 (Neurod6) and
Neurogranin (Nrgn), as statistically significant, and reliably clas-
sify these distinct cell types. However, below 100 transcripts per
cell, cell-type classification becomes inaccurate, and this is
correlated with markers such as Neurod6 being no longer statis-
tically associated with the first principal component.
We were able to reach similar conclusions with three other
single-cell mRNA-seq datasets (Shalek et al., 2013; Treutlein
et al., 2014; Kumar et al., 2014). With similarly low sequencing
depths, we were able to distinguish transcriptional states of sin-
gle cells collected across stages of the developing mouse lung
(Figures S2BS2D), wild-type mouse embryonic stem cells
from stem cells with a single gene knockout (Figures S2E
S2G), and heterogeneity within a population of bone-marrow-
derived dendritic cells (Figures S2HS2J). These results were
also not PCA-specific. We additionally examined two of these
datasets with t-distributed Stochastic Neighbor Embedding
(t-SNE) and Locally Linear Embedding (LLE), two nonlinear alter-
natives to PCA (Van der Maaten and Hinton, 2008; Roweis and
Saul, 2000) and achieved successful classification of transcrip-
tional states (Figures S2K and SKL), in each case recapitulating
the results of the original studies with fewer than 5,000 reads per
Cell Systems 2, 239250, April 27, 2016 243
Figure 3. Transcriptional States of Single Cells in the Mouse Brain Are Distinguishable at Low Transcript Coverage
(A) Principal component error as a function of read depth for selected principal components for the Zeisel et al. (2015) data.
(B) Accuracy of cell type classification as a function of transcripts per cell. Accuracy plateaus with increasing transcript coverage. At 1,000 transcripts per cell, all
three cell types can be distinguished with low error. At 100 transcripts per cell, pyramidal cells cannot be distinguished from each other, while oligodendrocytes
remain distinct.
(C) Covariance matrix of genes with high absolute loadings in the first principal component (left). The genes with the 100 highest positive and 100 lowest negative
loadings are displayed. The first principal component is enriched for genes indicative of oligodendrocytes and neurons (middle). Gene significance as a function of
transcript count for the first principal component (right).
(D) True and false detection rates as a function of transcript count for genes significantly associated with the first three principal components. Below 100
transcripts per cell, false positives are common.

cell. These results suggest that low dimensionality enables high
accuracy classification at low read depth across many methods.
Gene Expression Covariance Induces Tolerance to
Shallow Sequencing Noise
In the datasets we considered, the dominant noise-robust prin-
cipal components corresponded directly to large modules of
covarying genes. Such modules are common in gene expres-
sion data (Eisen et al., 1998; Alter et al., 2000; Bergmann
et al., 2003; Segal et al., 2003). We therefore studied the contri-
bution of modularity to principal component robustness in a
simple, mathematical model of gene expression (Supplemental
Information, section 2.2). Our analysis showed that the variance
explained by a principal component, and hence its noise toler-
ance, increases with the covariance of genes within the associ-
ated module (Figure 4A) and also the number of genes in the
module (Figures S3AS3C). While highly expressed genes
also contribute to noise tolerance, in the Shen et al. (2012) da-
taset we found little correlation between the expression level of
a gene and its contribution to the error of the first principal
component (R2 = 0.13; Figure S3D).
This analysis predicts that the large groups of tightly co-
varying genes observed in the Shen et al. (2012) and Zeisel
et al. (2015) datasets will contribute significantly to principal
value separation and noise tolerance. To directly quantify the
contribution of covariance to principal value separation in these
data, we randomly shuffled the sample labels for each gene. In
the shuffled data, genes vary independently, which eliminates

244 Cell Systems 2, 239250, April 27, 2016

Figure 4. Modularity of Gene Expression Enables Accurate, Low-Depth Transcriptional Program Identification
(A) Variance explained and covariance matrix for increasing gene expression covariance in a model.
(B) Variance explained by different principal components for the Zeisel et al. (2015) dataset. Covariance matrix shows large modules of covarying genes (middle).
Dominant transcriptional programs are robust to low-coverage profiling as predicted by model (bottom). Shuffling the dataset destroys the modular structure,
resulting in noise-sensitive transcriptional programs. For the shuffled data, 4,250 transcripts are required for 80% accuracy of the first three principal compo-
nents, whereas 340 transcripts suffices for the original dataset.

gene-gene covariance and raises the effective dimensionality
of the data. In contrast to the natural, low-dimensional data,
the principal values of the resulting data were nearly uniform
in magnitude. This significantly diminished the differences be-
tween the leading principal values within the shuffled data (Fig-
ure 4B, top).
Consequently, reconstruction of the principal components
became more read-depth intensive. For instance to recover
the first principal component with 80% accuracy from the shuf-
fled Zeisel et al. (2015) data, 12.5 times more transcripts are
required than for the unshuffled data (Figure 4B, bottom). We
reached a similar conclusion for the mouse ENCODE data,
where shuffling also decreased the differences between the
leading principal values and the rest, causing a 23-fold increase
in sequencing depth required to recover the first principal
component with 90% accuracy (Figure S4).
Large-Scale Survey Reveals that Shallow mRNA-Seq Is
Widely Applicable due to Gene-Gene Covariance
Both our analysis of Equation 1 and our computational investiga-
tions of mRNA-seq datasets suggest that high gene-gene co-
variances increase the distance of leading principal values
from the rest, thereby enabling the recovery of dominant prin-
cipal components at low mRNA-seq read depths. This finding,
if a common phenomenon, suggests that shallow mRNA-seq
may be rigorously employed when answering many biological
questions. To assess whether our findings are broadly appli-
cable, we performed a broad computational survey of available
gene expression data.
Since both gene covariances and principal values are funda-
mental properties of the biological systems under study, these
quantities may be analyzed using the wealth of microarray data-
sets available, leveraging a larger collection of gene expression
datasets as compared to mRNA-seq (see Figure S5A for ana-
lyses of several mRNA-seq datasets). We selected 352 gene
expression datasets from the GEO (Edgar et al., 2002) spanning
three species (yeast, 20 datasets; mouse, 106 datasets; and hu-
man, 226 datasets) that each contained at least 20 samples and
were performed on the Affymetrix platform.
Despite the differences between these datasets in terms of
species and collection conditions, they all possessed favorable
principal value distributions reflecting an effective low dimen-
sionality. For instance, on average the first principal value was
roughly twice as large as the second principal value, and
together the first five principal values explained a significant ma-
jority of the variance, suggesting that these datasets contain a
few, dominant principal components (Figure 5A, left). By shuf-
fling these datasets to reorder the sample labels for each gene,

Cell Systems 2, 239250, April 27, 2016 245

Figure 5. Gene Expression Survey of 352 Public Datasets Reveals Broad Tolerance of Bioinformatics Analysis to Shallow Profiling
(A) Variance explained by the first five transcriptional programs of 352 published yeast, mouse, and human microarray datasets (left). Shuffling microarray
datasets removes gene-gene covariance and destroys the relative dominance of the leading transcriptional programs. Read depth required to recover with 80%
accuracy the first five principal components of the 352 datasets (right). Removing gene expression covariance from the data requires a median of approximately
ten times more reads to achieve the same accuracy.
(B) Accuracy of GSEA of the human microarray datasets at low read depth (100,000 reads, i.e., 1% deep depth). Reactome pathway database gene sets are
correctly identified (blue) or not identified (yellow) at low read depth (false positives in red). !80% of gene sets can be correctly recovered at 100,000 reads.
(C) Accuracy of GSEA as a function of read depth.

we again found that these principal components emerge from
gene-gene covariance.
We related this pattern of dominant principal components
to the ability to recover biological information with shallow
mRNA-seq in these datasets. To generate synthetic mRNA-
seq data from these microarray datasets, we applied a probabi-
listic model to simulate mRNA-seq at a given read depth (see the
Experimental Procedures). We found that with only 60,000 reads
per sample, 84% of the 352 datasets have %20% error in their
first principal component. This translates into an average of
almost 1,000% read depth savings to recover the first principal
component with an acceptable PCA error tolerance of 20% (Fig-
ure 5A, right). By applying gene set enrichment analysis (GSEA)
to the first principal component of each of the 352 datasets at
low (100,000 reads per sample) and high read depths (10 million
reads per sample), we found that >60% of gene set enrichments
were retained with only 1% of the reads (Figures 5B and 5C). This
analysis demonstrates that biological information was also re-
tained at low depth.
Collectively, our analyses demonstrate that the success of
low-coverage sequencing relies on a few dominant transcrip-
tional programs. We also show that many gene expression data-
sets contain such noise-resistant programs as determined by
PCA and identified them with dominant dimensions in the data-
set. Furthermore, low dimensionality and noise robustness are
properties of the gene expression datasets themselves and exist
independent of the choice of analysis technique. Therefore, un-
supervised learning methods other than PCA would reach similar
conclusions, an expectation we verified using non-negative ma-
trix factorization (Figure S5B).

246 Cell Systems 2, 239250, April 27, 2016

The Read Depth Calculator: A Quantitative Framework
for Selecting Optimal mRNA-Seq Read Depth and
Number of Biological Samples
Because the optimal choice of read depth in an mRNA-seq
experiment is of widespread practical relevance, we developed
a read depth calculator that can provide quantitative guidelines
for shallow mRNA-seq experimental design. Having pinpointed
the factors that determine the applicability of shallow mRNA-
seq, we applied this understanding to determine the read depth
and number of biological samples to profile when designing an
experiment. To do so, we simplified the principal component er-
ror described by Equation 1 by assuming that the principal
values of mRNA-seq data are well separated, i.e., that the ratio experiment or less than 500,000 reads in total, occupying less
between consecutive principal values li+1/li is small (as defined
in the Supplemental Information, section 2.1), an assumption
justified by our large-scale microarray survey (see Figures S5C
and S5D). These assumptions enable us to provide simple
guidelines for making important experimental decisions, for
example, choosing read depth, N:
k2
Nz 2
(Equation 2)
nli kpci $ d
pci k
where n is the number of biological samples and k is a constant
that can be estimated from existing data (see the Supplemental
Information, section 2.1 for a derivation of this equation and its
limitations). This relationship can be understood intuitively. First,
Equation 2 states that the principal component error decreases
with read depth, a consequence of the well-known fact that the
signal-to-noise
p ratio of a Poisson random variable is proportional
to N. The read depth also depends on li, which comes from the
li $ lj term of Equation 1. Finally, the influence of the sample
number n on read depth follows from the definition of covariance
as an average over samples. (Figure S5E shows that n is approx-
imately statistically uncorrelated with principal values across the
microarray datasets.)
Equation 2 has implications for optimizing the tradeoff between
read depth and sample number in single-cell mRNA-seq experi-
ments. As principal component error depends on the product of
read depth and number of samples, error in mRNA-seq analyses
can be reduced equivalently in two ways, by either increasing the
total number of profiled cells or the transcript coverage. To illus-
trate this point, we computationally determined the error in the
first principal component of the single cell mouse brain data
from Zeisel et al. (2015) as a function of cell number. Consistent
with Equation 2, our calculations show that increasing the num-
ber of profiled cells reduces error in the first principal component
(Figure 6A). Furthermore, we show that with the Zeisel et al. (2015)
data, multiple different experimental configurations with the
same total number of transcripts can yield the same principal
component error. For example, 100,000 transcripts divided
between either 50 or 400 cells both yield a principal component
error of !20%. This result is of particular relevance in single-
cell experiments because transcript depth per cell is currently
limited by a !20% mRNA capture efficiency, and so cannot be
easily increased (Shalek et al., 2014). In such cases, limited
sequencing resources might be best used to sequence more
cells at low depth rather than allocating sequencing resources
to oversampling a few thousand unique transcripts.
Experimentalists can use the read depth calculator to pre-
dict requirements for read depth or sample number in high-
throughput transcriptional profiling given their desired accuracy
based on the statistics of principal value separation in our global
survey. Figure 6B shows the reads required for desired accu-
racies and an assumed principal value for a human transcrip-
tional experiment with 100 samples (typical values for the first
five principal values for human are indicated in dashed lines).
As an illustration, a hypothetical experiment with a typical first
principal value of 1.4 3 10$5 (median principal value from the
226 human microarray datasets) and 100 samples where 80%
PCA accuracy is tolerable requires less than 5,000 reads per
than 0.125% of a single sequencing lane in the Illumina HiSeq
4000.
The predictions from this analytically derived read depth
calculator are demonstrably accurate. We compared the analyt-
ically predicted number of reads required for 80% PCA accuracy
in the first five transcriptional programs to the value determined
through simulated shallow mRNA-seq for 226 microarray and 4
mRNA-seq human datasets. We determined k empirically by
fitting 50% of the datasets. Cross-validation with the remaining
50% of the datasets showed remarkable agreement between
the analytical predictions and computationally determined
values. In these calculations, the analytically predicted number
of reads required to reach 80% accuracy deviates from the
depth required in simulation by less than 10% (Figure 6C). The
read depth calculator is available online (http://thomsonlab.
github.io/html/formula.html).
Finally, while we use the first principal component for illustra-
tion, Equation 2 can be applied to any principal component,
including the trailing principal components. Recent work dis-
cusses a statistical method to identify those principal compo-
nents that are likely to be informative, and this work can be
used in conjunction with Equation 2 to pinpoint the relevant prin-
cipal components and the sequencing parameters needed to es-
timate them satisfactorily (Klein et al., 2015).
DISCUSSION
Single-cell transcriptional profiling is a technology that holds
the promise of unlocking the inner workings of cells and uncov-
ering the roots of their individuality (Klein et al., 2015; Macosko
et al., 2015). We show that for many applications that rely on
the determination of transcriptional programs, biological in-
sights can be recapitulated at a fraction of the widely proposed
high read depths. Our results are based on a rigorous mathe-
matical framework that quantifies the tradeoff between read
depth and accuracy of transcriptional program identification.
Our analytical results pinpoint gene-gene covariance, a ubiqui-
tous biological property, as the key feature that enables un-
compromised performance of unsupervised gene expression
analysis at low read depth. The same mathematical framework
also leads to practical methods to determine the optimal
read depth and sample number for the design of mRNA-seq
experiments.
Given the principal values that we observe in the human micro-
array datasets, our analysis suggests that one can profile tens of
thousands of samples, as opposed to dozens, while still being
Cell Systems 2, 239250, April 27, 2016 247
Figure 6. Mathematical Framework Provides a Read Depth Calculator and Guidelines for Shallow mRNA-Seq Experimental Design
(A) Error in the first principal component of the Zeisel et al. (2015) dataset for varying cell number and read-depth. Black circles denote a fixed number of total
transcripts (100,000). Error can be reduced by either increasing transcript coverage or the number of cells profiled.
(B) Number of reads required (color) to achieve a desired error (y axis) for a given principal value (x axis). Typical principal values (dashed black vertical lines) are
the medians across the 352 gene expression datasets.
(C) Error of the read depth calculator (Equation 2) across 176 gene expression datasets used for validation (out of 352 total). The calculator predicts the number of
reads to achieve 80% PCA accuracy in each dataset (colored dots). The predicted values closely agree with simulated results, with the median error <10% for the
first five transcriptional programs.

able to accurately identify transcriptional programs. At this scale,
researchers can perform entire chemical or genetic knockout
screens or profile all !1,000 cells in an entire Caenorhabditis
elegans, 40 times over, in a single 400,000,000 read lane on
the Illumina HiSeq 4000. Because shallow mRNA-based screens
would provide information at the level of transcriptional pro-
grams and not individual genes, complementing these experi-
ments by careful profiling of specific genes with targeted
mRNA-seq (Fan et al., 2015) or samples of interest with conven-
tional deep sequencing would provide a more complete picture
of the relevant biology.
Fundamentally, our results rely on a natural property of gene
expression data: its effective low dimensionality. We observed
that gene expression datasets often have principal values that
span orders of magnitude independently of the measurement
platform and that this property is responsible for the noise toler-
ance of early principal components. These leading, noise-robust
principal components are effectively a small number of dimen-
sions that dominate the biological phenomena under investiga-
tion. These insights are consistent with previous observations
that were made following the advent of microarray technology
(Eisen et al., 1998; Segal et al., 2003; Bergmann et al., 2003), pro-
posing that low dimensionality arises from extensive covariation
in gene expression. We suggest that the covariances and prin-
cipal values in gene expression are determined by the architec-
tural properties of the underlying transcriptional networks, such
as the co-regulation of genes, and therefore it is the biological
system itself that confers noise tolerance in shallow mRNA-seq
measurements. Related work in neuroscience has explored the
implications of hierarchical network architecture for learning
the dominant dimensions of data (Saxe et al., 2013; Hinton and
Salakhutdinov, 2006).

248 Cell Systems 2, 239250, April 27, 2016

 Discovering and exploiting low dimensionality to reduce un-
certainty in measurements is at the heart of modern signal pro-
cessing techniques (Donoho 2006; Candes et al., 2006). These
methods first found success in imaging applications, where
low dimensionality arises from the statistics and redundancies
of natural images, enabling most images to be accurately repre-
sented by a small number of wavelets or other basis functions.
Our results suggest that shallow mRNA-seq is similarly enabled
by an inherent low dimensionality in gene expression datasets
that emerges from groups of covarying genes. Just as only a
few wavelets are needed to represent most images, only a few
groups of transcriptional programs seem to be necessary to pro-
duce a coarse-grained representation of transcriptional state.
We believe that the measurement of many diverse biological
systems could benefit from the identification and analysis of hid-
den low-dimensional representations. For instance, proteome
quantification, protein-protein interactions, and human genetic
variant data all contain high levels of correlations, suggesting
these datasets may all be effectively low dimensional. We antic-
ipate new modes of biological inquiry as advances from signal
processing are integrated into biological data analysis and as
the underlying structural features of biological networks are ex-
ploited for large-scale measurements.
EXPERIMENTAL PROCEDURES
Simulated Shallow Sequencing through Down-sampling of Reads
Transcriptional datasets were obtained from the GEO (Zeisel et al. [2015] was
from http://www.linnarssonlab.org). mRNA-seq read counts were normalized
by the total number of reads in the sample. For each read depth, we model the
sequencing noise with a multinomial distribution. The Zeisel et al. (2015) data
were sampled without replacement because of the unique molecular identi-
fiers (see Supplemental Experimental Procedures).
Finding Genes Significantly Associated with a Principal Component
We first generated a null distribution of gene loadings from the principal com-
ponents of a shuffled, transcript-count matrix. All p values were computed with
respect to this distribution; averages over 15 replicates are reported.
Gene Set Enrichment Analysis
GSEA was performed with 1,370 gene lists from MSigDB (Subramanian et al.,
2005). The loadings of each principal component were collected in a distribu-
tion and loadings within 2 SDs from the mean of this distribution were consid-
ered for analysis. We applied a hypergeometric test with a significance p value
cutoff of 10$4.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures
and five figures and can be found with this article online at http://dx.doi.org/
10.1016/j.cels.2016.04.001.
AUTHOR CONTRIBUTIONS
G.H., H.E.-S., and M.T. conceived the idea. G.H. wrote the simulations
and analyzed data, with input from M.T. and H.E.-S. R.B. and M.T. performed
theoretical analysis. R.B. wrote the mathematical proofs. The manuscript was
written by G.H., R.B., H.E.-S., and M.T.
ACKNOWLEDGMENTS
The authors would like to thank Jason Kreisberg, Alex Fields, David Sivak, Pat-
rick Cahan, Jonathan Weissman, Chun Ye, Michael Chevalier, Satwik Ra-
jaram, and Steve Altschuler for careful reading of the manuscript; Eric Chow,
John Haliburton, Sisi Chen, and Emeric Charles for their experimental insights;
and Paul Rivaud for website design assistance. This work was supported
by the UCSF Center for Systems and Synthetic Biology (NIGMS P50
GM081879). H.E.S. acknowledges support from the Paul G. Allen Family Foun-
dation. M.T. acknowledges support from the NIH Office of the Director, the Na-
tional Cancer Institute, and the National Institute of Dental and Craniofacial
Research (NIH DP5 OD012194).
Received: November 30, 2015
Revised: March 8, 2016
Accepted: April 4, 2016
Published: April 27, 2016
REFERENCES
Alon, U., Barkai, N., Notterman, D.A., Gish, K., Ybarra, S., Mack, D., and
Levine, A.J. (1999). Broad patterns of gene expression revealed by clustering
analysis of tumor and normal colon tissues probed by oligonucleotide arrays.
Proc. Natl. Acad. Sci. USA 96, 67456750.
Alter, O., Brown, P.O., and Botstein, D. (2000). Singular value decomposition
for genome-wide expression data processing and modeling. Proc. Natl.
Acad. Sci. USA 97, 1010110106.
Bengio, Y., Delalleau, O., Le Roux, N., Paiement, J.-F., Vincent, P., and
Ouimet, M. (2004). Learning eigenfunctions links spectral embedding and
kernel PCA. Neural Comput. 16, 21972219.
Bergmann, S., Ihmels, J., and Barkai, N. (2003). Iterative signature algorithm
for the analysis of large-scale gene expression data. Phys. Rev. E Stat.
Nonlin. Soft Matter Phys. 67, 031902.
Bonneau, R. (2008). Learning biological networks: from modules to dynamics.
Nat. Chem. Biol. 4, 658664.
Candes, E.J., Romberg, J.K., and Tao, T. (2006). Stable signal recovery from
incomplete and inaccurate measurements. Commun. Pure Appl. Math. 59,
12071223.
Ding, C., and He, X. (2004). K-means clustering via principal component anal-
ysis. ICML Proceedings of the 21st International Conference on Machine
Learning (ACM), p. 29.
Donoho, D.L. (2006). Compressed sensing. IEEE Trans. Inf. Theory 52, 1289
1306.
Duarte, M.F., Davenport, M.A., Takbar, D., Laska, J.N., Sun, T., Kelly, K.F., and
Baraniuk, R.G. (2008). Single-pixel imaging via compressive sampling. IEEE
Signal Process. Mag. 25, 8391.
Edgar, R., Domrachev, M., and Lash, A.E. (2002). Gene Expression Omnibus:
NCBI gene expression and hybridization array data repository. Nucleic Acids
Res. 30, 207210.
Eisen, M.B., Spellman, P.T., Brown, P.O., and Botstein, D. (1998). Cluster anal-
ysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci.
USA 95, 1486314868.
Fan, H.C., Fu, G.K., and Fodor, S.P.A. (2015). Expression profiling.
Combinatorial labeling of single cells for gene expression cytometry. Science
347, 1258367.
Ham, J., Lee, D.D., Mika, S., and Scholkopf, B. (2004). A kernel view of the
dimensionality reduction of manifolds. ICML Proceedings of the 21st
International Conference on Machine Learning (ACM), p. 47.
Hinton, G.E., and Salakhutdinov, R.R. (2006). Reducing the dimensionality of
data with neural networks. Science 313, 504507.
Holter, N.S., Maritan, A., Cieplak, M., Fedoroff, N.V., and Banavar, J.R. (2001).
Dynamic modeling of gene expression data. Proc. Natl. Acad. Sci. USA 98,
16931698.
Jaitin, D.A., Kenigsberg, E., Keren-Shaul, H., Elefant, N., Paul, F., Zaretsky, I.,
Mildner, A., Cohen, N., Jung, S., Tanay, A., and Amit, I. (2014). Massively par-
allel single-cell RNA-seq for marker-free decomposition of tissues into cell
types. Science 343, 776779.
Cell Systems 2, 239250, April 27, 2016 249
Klein, A.M., Mazutis, L., Akartuna, I., Tallapragada, N., Veres, A., Li, V.,
Peshkin, L., Weitz, D.A., and Kirschner, M.W. (2015). Droplet barcoding for sin-
gle-cell transcriptomics applied to embryonic stem cells. Cell 161, 11871201.
Kliebenstein, D.J. (2012). Exploring the shallow end; estimating information
content in transcriptomics studies. Front. Plant Sci. 3, 213.
Kumar, R.M., Cahan, P., Shalek, A.K., Satija, R., DaleyKeyser, A.J., Li, H., Zhang,
J., Pardee, K., Gennert, D., Trombetta, J.J., et al. (2014). Deconstructing tran-
scriptional heterogeneity in pluripotent stem cells. Nature 516, 5661.
Macosko, E.Z., Basu, A., Satija, R., Nemesh, J., Shekhar, K., Goldman, M.,
Tirosh, I., Bialas, A.R., Kamitaki, N., Martersteck, E.M., et al. (2015). Highly par-
allel genome-wide expression profiling of individual cells using nanoliter drop-
lets. Cell 161, 12021214.
Ng, A.Y., Jordan, M.I., and Weiss, Y. (2001). On spectral clustering: analysis
and an algorithm. In Advances in Neural Information Processing Systems
(MIT Press), pp. 849856.
Patel, A.P., Tirosh, I., Trombetta, J.J., Shalek, A.K., Gillespie, S.M., Wakimoto,
H., Cahill, D.P., Nahed, B.V., Curry, W.T., Martuza, R.L., et al. (2014). Single-
cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma.
Science 344, 13961401.
Pollen, A.A., Nowakowski, T.J., Shuga, J., Wang, X., Leyrat, A.A., Lui, J.H., Li,
N., Szpankowski, L., Fowler, B., Chen, P., et al. (2014). Low-coverage single-
cell mRNA sequencing reveals cellular heterogeneity and activated signaling
pathways in developing cerebral cortex. Nat. Biotechnol. 32, 10531058.
Ringner, M. (2008). What is principal component analysis? Nat. Biotechnol. 26,
303304.
Roweis, S.T., and Saul, L.K. (2000). Nonlinear dimensionality reduction by
locally linear embedding. Science 290, 23232326.
Saxe, A.M., Mcclelland, J.L., and Ganguli, S. (2013). Learning hierarchical
category structure in deep neural networks. Proceedings of the 35th Annual
Meeting of the Cognitive Science Society, pp. 12711276.
Segal, E., Shapira, M., Regev, A., Peer, D., Botstein, D., Koller, D., and
Friedman, N. (2003). Module networks: identifying regulatory modules and
their condition-specific regulators from gene expression data. Nat. Genet.
34, 166176.
Shai, R., Shi, T., Kremen, T.J., Horvath, S., Liau, L.M., Cloughesy, T.F.,
Mischel, P.S., and Nelson, S.F. (2003). Gene expression profiling identifies mo-
lecular subtypes of gliomas. Oncogene 22, 49184923.
250 Cell Systems 2, 239250, April 27, 2016
Shalek, A.K., Satija, R., Adiconis, X., Gertner, R.S., Gaublomme, J.T.,
Raychowdhury, R., Schwartz, S., Yosef, N., Malboeuf, C., Lu, D., et al.
(2013). Single-cell transcriptomics reveals bimodality in expression and
splicing in immune cells. Nature 498, 236240, advance online publication.
Shalek, A.K., Satija, R., Shuga, J., Trombetta, J.J., Gennert, D., Lu, D., Chen,
P., Gertner, R.S., Gaublomme, J.T., Yosef, N., et al. (2014). Single-cell RNA-
seq reveals dynamic paracrine control of cellular variation. Nature 510,
363369.
Shankar, R. (2012). Principles of Quantum Mechanics (Springer Science &
Business Media).
Shen, Y., Yue, F., McCleary, D.F., Ye, Z., Edsall, L., Kuan, S., Wagner, U.,
Dixon, J., Lee, L., Lobanenkov, V.V., and Ren, B. (2012). A map of the cis-reg-
ulatory sequences in the mouse genome. Nature 488, 116120.
Stewart, G.W., and Sun, J. (1990). Matrix Perturbation Theory (Academic
Press).
Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gillette,
M.A., Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S., and Mesirov, J.P.
(2005). Gene set enrichment analysis: a knowledge-based approach for inter-
preting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA 102,
1554515550.
Trapnell, C. (2015). Defining cell types and states with single-cell genomics.
Genome Res. 25, 14911498.
Treutlein, B., Brownfield, D.G., Wu, A.R., Neff, N.F., Mantalas, G.L., Espinoza,
F.H., Desai, T.J., Krasnow, M.A., and Quake, S.R. (2014). Reconstructing line-
age hierarchies of the distal lung epithelium using single-cell RNA-seq. Nature
509, 271375.
Van der Maaten, L., and Hinton, G. (2008). Visualizing data using t-SNE.
J. Mach. Learn. Res. 9, 85.
Velculescu, V.E., Madden, S.L., Zhang, L., Lash, A.E., Yu, J., Rago, C., Lal, A.,
Wang, C.J., Beaudry, G.A., Ciriello, K.M., et al. (1999). Analysis of human tran-
scriptomes. Nat. Genet. 23, 387388.
Zeisel, A., Munoz-Manchado, A.B., Codeluppi, S., Lonnerberg, P., La Manno,
G., Jureus, A., Marques, S., Munguba, H., He, L., Betsholtz, C., et al. (2015).
Brain structure. Cell types in the mouse cortex and hippocampus revealed
by single-cell RNA-seq. Science 347, 11381142.
Cell Systems, Volume 2

Supplemental Information

Low Dimensionality in Gene Expression Data

Enables the Accurate Extraction of Transcriptional
Programs from Shallow Sequencing

Graham Heimberg, Rajat Bhatnagar, Hana El-Samad, and Matt Thomson

 SUPPLEMENTAL INFORMATION
Table of Contents

Supplemental Figures
Figure S1
Figure S2
Figure S3
Figure S4
Figure S5

1 Supplemental Figure Legends 1

2 Supplemental Theory 4
2.1 Shallow sequencing 4
2.2 Gene expression modules 12
3 Supplemental Experimental Procedures 14
Supplemental References 17
 100%
A Principal component error
B
Program 1
80% 2
3 80%

Principal component error

8
15
60%
60% 10

40% 32
100
40%
316
20% 1,000
3,162
20%
3 4 5 6 7 10,000 (x103 reads)
10 10 10 10 10
Reads per sample 10 20 30
Principal component index
C D
1
principal component 1

Negative Predictive Value

formation of tubulin folding intermediates...
prefoldin mediated transfer of substrate... Significance
-log10(p)
sig regulation of the actin cytoskeleton...
10
axon guidance
protein folding 0.99
cell cycle
4 Principal component 1
mitotic prometaphase
Insignificant 2
pathogenic escherichia coli infection
0 3
oocyte meiosis
ncadherin pathway
1 10 100 1000 10000
0.98
3
Reads per sample(10 ) 3 4 5 6 7
10 10 10 10 10
Reads per sample
E 1 F
100%
Principal component error in FPKM units
Positive Predictive Value

principal component error

0.9 80% Principal component 1

2
3
8
15
Principal component 1 60%
0.8 2
3
40%

0.7 1% deep 99%

20% read depth
3 4 5 6 7
10 10 10 10 10
Reads per sample 3 4 5 6 7
10 10 10 10 10
Reads per sample
G H
Distinguishing tissue types with FPKM units 10,000 reads
5
Principal Value Decay in FPKM units x 10 x 10
3
32,000 reads
Differences between principal values (+1)

8
70,000 reads
20% 1
bonemarrow
Variance explained

liver
4 spleen
E14.5liver
thymus
10% 0.5
cerebellum
PC3

0
cortex
olfactory
4 MEF
mESC
E14.5brain
4 8 12 16
program index (i) E14.5limb
8
heart
10 5 0 5 E14.5heart
PC2 x 10
3

Figure SI 1
 A

Differences between principal values Differences between principal values

Variance explained (Zeisel et al.)

5
x 10

20%
5

5 10 15 20
B principal value index C D
Variance explained (Truetlein et al.)

5 4
x 10 100% x 10
8%

Principal component error

4
1.2 80%

Principal component 2
6% PC 1 0
2
60% 3
0.8
4 3200 reads
4% 40% 6800 reads
1% deep
read depth 99% 8 15,000 reads
0.4
20% Day 14.5 (1107 reads)
2% Day 16.5 (1107 reads)
1
Day 18.5 (1107 reads)
10
3
10
4
10
5
10
6 7
10 8 4 0 4 8
5 10 15 20 Principal component 1
principal value index read depth x 10
4

E F G
variance explained (Kumar et al.)

5 4
x 10 100% x 10
Differences between principal values

Principal component error

Principal component 2
1 3200 reads
30% 1.2 80%
6800 reads

60% 0 32,000 reads

20% 0.8 PC 1
2 Dgrc8 -/-
3 1
40% WT
1% deep
10% 0.4 read depth 99% Neural Progenitor
20%

5 10 15 20 3 4 5 6
10 10 10 10 10
7
0 2 4 4
principal value index read depth Principal component 1
x 10

H I J
Differences between principal values

3
x 10
5 100% x 10
Variance explained (Shalek et al.)

20% 2
Principal component error

Classification line
2200 reads
Principal component 2

80% 2
4600 reads
10,000 reads
60% PC 1 1
2 1107 reads
3
10% 1
40%
0

1% deep 99%
20%
read depth 1
not mature mature
3 4 5 6 7
4 8 12 16
10 10 10 10 10 1 0 1 2 3
principal value index read depth Principal component 1 x 10
3

t-SNE
Data from Kumar et al.

LLE

t-SNE
Data from Shalek et al.

LLE

Figure SI 2
 A 1 B pc1
5 1

gene expression
1 between cluster
cluster
i 300 covariance r
1
1 within cluster
covariance
100
High
covariance

No
covariance
gene variance mi
Negative
covariance

D
C 4
10
R2=0.13
Principal value magnitude

PC 1 loading error (%)

20 1
2
10
1
1

10 0
10

10
Increasing within cluster covariance
High
covariance

5 4 3
genes

No 10 10 10 10
covariance
Gene expression level (normalized read counts)
genes Negative
covariance

Figure SI 3
 20%
variance explained

u ns
hu
10%
d
e da
ta
shu ed data

10 20 30
principal value index
Modular Non-modular High
(covariance removed) covariance

genes
genes

Negative
covariance
genes genes

Noise Tolerant Noise Sensitive

~100,000 reads ~ 2.3 million reads
100%
100%
Principal component error

Principal component error

90% accuracy

80% 80% 90% accuracy

60% 60%

40% 40% Program 1

2
3
20% Program 1 20%
2
3
3 4 5 6 7 3 4 5 6 7
10 10 10 10 10 10 10 10 10 10
Reads per sample Reads per sample

data: Shen et al.

Figure SI 4
 A
Shen
Treutlein
5 Shalek
10 Kumar
Chen
Pollen
Median (shuffled)

i - i+1

10

10

8
10
2 3 4 5 6 7 8 9
principal value index

B
Yeast Mouse Human

0
R2= 0.19 10
median NMF error R2= 0.69 R2= 0.48
median NMF error

median NMF error

1
210

1
1 10
10
1
10
5 6 5 4 5 4
510 10
4
10 10 10 10 10
1 2 3 1 2 3 1 2 3

C Principal value decay human data D Principal value decay shuffled human data

1 1
i/ 1

i/ 1

0.5 0.5

2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10
principal value index principal value index (shuffled data)

E
3 3 3
10 10 10 Human
Yeast Mouse
R2= 0.0846 R2= 0.0007
first principal value (1

first principal value (1

first principal value (1

R2= 0.0426

5 5
10 10
5
10

1 2 10 1 2 10 1 2 3
10 10 10 10 10 10 10
sample number (n) sample number (n) sample number (n)

Figure SI 5